A dynamic recommendation system design method based on multi-dimensional classification reinforcement learning

Abstract

The invention relates to a dynamic recommendation system design method based on multi-dimensional classification reinforcement learning, and belongs to the technical field of reinforcement learning and recommendation. The method includes: 1, classifying all items according to the inherent attributes of recommended items, performing heat statistics on all items, and updating representative items; 2, that user sends a request to the server and requests the server to recommend the article; 3, calculating a user activity degree and a network weight and storing that activity degree and the networkweight; 4, that server judges wheter to skip to step 5 according to the activity degree of the user; 5, that server recommends an item to the us according to the actor neural network and the existinguser state vector; 6, updating that status of the user if the us carries out feedback on the items recommended by the server, and returning to the step 3; Otherwise no action. The invention more objectively reflects the relationship between the user articles and the change of the user interest. The accuracy of the recommendation is enhanced by user activity and static information registered by theuser.

Description

technical field [0001] The invention relates to a design method of a dynamic recommendation system based on multidimensional classification reinforcement learning, which belongs to the technical field of reinforcement learning and recommendation. Background technique [0002] In recent years, the development of the Internet has brought about an explosive growth in the amount of information. In application scenarios such as search engines, e-commerce, and news feeds, in order to enable users to see interesting information in a short time, recommendation systems play an increasingly important role. Traditional recommendation systems mainly use: collaborative filtering-based methods, content-based methods, and hybrid methods; in recent years, as an extension of traditional methods, deep learning models have been introduced. These models are considered to better model the connection between users and items. Although they have improved compared to traditional algorithms, they st...

AI Technical Summary

Problems solved by technology

[0004]2. The dynamic characteristics of the recommendation system have not been well resolved

At present, there is a lack of an overall design of a recommender system that can rationally utilize item features and simultaneously solve the two main problems mentioned above

Images